Issue Tree

Where it came from

McKinsey, and specifically Barbara Minto, who joined the firm in 1963 as one of its first women consultants and was soon asked to teach the others how to write. What she taught was a way of organising thought: any question can be broken into a small number of parts, and a breakdown is only trustworthy if the parts are mutually exclusive and collectively exhaustive. No overlaps, no gaps. The consulting world still says MECE, pronounced "mee-see", and still means her rule.

Her book The Pyramid Principle came out in 1978 and has never gone out of print. The issue tree is the working end of it: the question at the top, the parts beneath, each part split again until the leaves are things you can check. The other half of her book, how to present the answer once you have it, is the method Greggle calls Answer First.

The habit of writing your best answer before you start comes from the same firm. A tree built without a hypothesis finds whatever it was pointed at, and takes far longer doing it.

Why it works

A question split into non-overlapping parts can be worked on in pieces, by different people, in any order. The "nothing left out" test catches the branch you forgot about before it becomes the reason the plan failed. And a branch that says what would have to be true, and how you would test it, can be closed. Most questions stay open for years because nobody ever wrote down what would settle them.

Greggle is built in this shape. Every board is a split, and every step is a branch you can open and split again. The issue tree is the method that says out loud what the app does quietly, and adds the two questions the app can’t ask for you: does this branch overlap another, and would it change the answer?

When to reach for it, and when not to

Reach for it when a question is too big to see whole, when several people disagree about it and you suspect they are answering different parts, or when you want to know what to check first. It suits a Challenge, a Problem that has many possible causes, and a Decision where the options haven’t yet been separated from the reasons.

It is the wrong tool when the question is really a chain of causes, which is 5 Whys, or when the answer depends mostly on how the future turns out, which is Scenario Planning. Greggle suggests it on a Challenge, a Problem and a Decision.

How to run it

With a pen, on one sheet of paper.

Write one question, singular and answerable. Then write your best answer today, before you start, and who decides what once it is answered. If nobody would do anything differently, swap the question for one where they would.
Split the question into two to six parts. Test the split: could a fact belong to two parts? Is there a fact that belongs to none? Write down what the split leaves out, and why you cut it this way rather than the obvious other way.
For each part, say what would have to be true for it to hold, stated so that it could turn out false, and how you would actually test it. "More analysis" is not a test.
Judge each part: if it holds, does it settle the question, narrow it, or barely move it? Most parts are not decisive. If you have marked them all decisive, you have marked none.
Open the parts that would change the answer and split them again. Leave the others whole. Depth on the interesting branch instead of the decisive one is the commonest failure.
Pick the test to run first: the one that moves the answer furthest for the least effort. Give it a date and write down what result would change your mind.

There is a one-page card with these steps and a blank grid to fill in, made to print on A4. Or the PDF.

On a Greggle board

Pick Issue Tree on a Challenge, Problem or Decision board and Greggle proposes the question, the first split with three branches, and what to test first. The depth comes from you: open any branch and split it again.

Should we run the stall again in December, inside the split: three branches that do not overlap, the first with its test and its verdict — a real board · the December stall, inside the split · click a step to open it

Inside 2 — The split: the board one level down — a real board · the December stall, inside the split · click a step to open it

1The questionAnswerable and singular. Your best answer today, said before you start. The decision this informs.

2The splitWhat this split leaves out. Why cut it this way. Then the branches: three to start, two to six.

each branchThis part of the questionDoes it overlap another, and if so which one owns it? What would have to be true? How you would test it? If it holds: settles, narrows, or barely moves the question.

open a branchSplit it againAny branch becomes a board of its own. Go deep on the decisive ones and leave the rest whole.

3What to test firstThe test to run first. By when. What result would change your mind.

not on the boardMECEThe app can’t judge overlap for you. The two questions on each branch are how you do it yourself.

Nothing is added until you have looked at it. Open any step and it becomes a board of its own. Steps that cannot sensibly be cut smaller get marked as actions, and every action collects in one flat list.

What the check looks for

One button reads the whole branch back. On this method it reports, among other things:

Branches have been written and not one says whether it overlaps another. Overlap does the most damage here, because the same thing gets counted twice.
Not one branch says how it would be tested. A tree made of untestable branches organises the question without ever answering it.
None of the branches would settle the question, or all of them carry the same weight, so the tree cannot say what to do first.
One branch has been taken several levels down while another has not been opened at all.

A worked example

a Decision · should we run the market stall again in December?

The questionShould the three of us rent the stall again for the four Saturdays in December? Best answer today: yes. Decision it informs: whether to pay the December fee, which is due on 20 November.

The splitDid November actually make money, once everything is counted? Would December be different from November? Can the three of us do it in December? Left out: whether we would rather do something else with four Saturdays. Why this cut: it separates facts we can check now from guesses about next month from a conversation the three of us need to have.

Branch 1Separate. Would have to be true: takings minus ingredients minus the fee was more than we would have earned doing anything else. Test: the takings sheet and the receipts, added up honestly. If it fails, it settles the question.

Branch 2Separate. Would have to be true: December has more shoppers and no new stall selling the same thing. Test: ask the market manager how many stalls are booked and what last December’s footfall was. Narrows it.

Branch 3Overlaps branch 1 slightly, since the 6am problem cost us stock in November; branch 1 owns that. Would have to be true: nobody has mocks on a December Saturday and the 6am person is still willing. Test: the mocks timetable and one conversation. Settles it if it fails.

Test firstCount November properly, by Tuesday. What would change my mind: if the money after the fee is less than three Saturdays of babysitting, the answer is no, however good December looks.

The best answer was yes. The first test was the one most likely to make it no, which is the right test to run first. Notice that branch 3 turned out to overlap branch 1, and saying which owns the overlap is what stops the 6am problem being counted twice.

Where it falls down

Perfect MECE is rarely achievable and chasing it wastes time. Good enough is "I can see no overlap and I can’t think of a gap", written down, with what the split leaves out named honestly. A split that claims to leave out nothing without having checked is worse than one that admits what it skipped.

The other failure is depth in the wrong place. The interesting branch gets four levels; the decisive branch gets none. The judgement on each branch, settles, narrows or barely moves, is there to stop that, and it only works if most branches are honestly marked as not decisive.

Pairs well with

Answer First. The other half of Minto’s book. The tree is how you work the answer out; Answer First is how you deliver it.
Cause and Effect. When the question is "why did this happen", the fishbone is an issue tree with the first split already chosen for you.
First Principles. When you can’t find a split that isn’t inherited from how everyone else frames the question, strip the assumptions first.